融合超像素分割与码本模型的目标检测算法  

Object Detection Algorithm Based on the Combination of the Superpixel Segmentation and Codebook Model

作　　者：凡时财[1] 曾祥峰[1] 周雪[1] 邹见效[1] 徐红兵[1] 

机构地区：[1]电子科技大学自动化工程学院,电子科技大学信息生物学研究中心,成都611731

出　　处：《电子科技大学学报》2017年第4期572-578,共7页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金面上项目(61472063);国家自然科学基金青年基金(61503061);中央高校基本科研业务费(ZYGX2016J102);吉林大学符号计算与知识工程教育部重点实验室开放基金(93K172017K02)

摘　　要：针对码本模型在前景目标检测中的效率有待进一步提高的现状,提出了融合超像素分割的码本构建算法。为减小处理对象的规模,设计了按照颜色及空间相似度聚类原始像素点的思路。以超像素作为码本构建单元,有利于抑制局部噪声并降低码本的冗余度。实验结果表明,融合超像素分割的码本模型算法在保持前景目标检测准确性的情况下,能显著减少视频处理过程中的内存消耗以及提高视频帧处理效率,在基于DM6437的嵌入式处理平台上达到了实时处理的性能。A novel codebook model combined with the superpixel segmentation method is proposed in this paper to improve the efficiency of object detection. The original pixels are clustered based on the similarities of both color and location information to reduce the processing cost. Our revised codebook model based on the superpixel could not only suppress the effect of local noise, but also reduce the redundancy of codebook. Simulation results indicate that, our proposed algorithm could reduce the memory consumption and improve the processing speed significantly without sacrificing the detection precision. Our algorithm could implement the object detection in real time in the embedded video processing system based on the DM6437 processor.

关 键 词：码本模型 检测效率 嵌入式处理平台 目标检测 超像素分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]